Trust Between AI Agents: Measuring Formation, Breakage, and Recovery, with Implications for Governing Multi-Agent Systems

The paper introduces a behavioral measure for quantifying trust among AI agents in cooperative environments, specifically through costly verification in a survival game. It analyzes trust dynamics using six model snapshots, including Claude Opus 4.6, Claude Sonnet 4.6, GPT-5.1, and Gemini 3.1 Pro, revealing that trusted agents significantly reduce verification efforts by 60-85%. The findings emphasize the importance of calibrating trust levels in multi-agent systems, as models exhibiting higher trust can make quicker decisions and achieve better outcomes, while over-verification leads to indecision and inefficiency.